高光谱成像技术结合特征提取方法的草莓可溶性固形物检测研究

采用高光谱成像技术结合不同的特征提取方法，实现了对草莓可溶性固形物含量的检测。通过提取154颗成熟无损伤草莓的高光谱图像的874～1 734 nm范围光谱信息，对941～1 612 nm光谱采用移动平均法(moving average，MA)进行预处理。基于残差法剔除19个异常样本后将剩余135个样本分为建模集(n=90)和预测集(n=45)。采用连续投影算法(successive projections algorithm, SPA)，遗传偏最小二乘算法(genetic algorithm-partial least squares, GAPLS)结合连续投影算法(GAPLS-SPA)，加权回归系数(weighted regression coefficient, Bw)以及CARS法(competitive adaptive reweighted sampling)选择特征波长分别提取14，17，24与25个特征波长，并采用主成分分析(principal component analysis, PCA)与小波变换(wavelet transform, WT)分别提取20与58个特征信息。分别基于全波段光谱、特征波长与特征信息建立PLS模型。所有模型都取得了较好的效果，基于全波段光谱的PLS模型与基于WT提取的特征信息的PLS模型的效果最优，建模集相关系数(r_c)与预测集相关系数(r_p)均高于0.9。结果表明高光谱成像技术结合特征提取方法可用于草莓可溶性固形物含量的检测。

Determination of Soluble Solid Content in Strawberry Using Hyperspectral Imaging Combined with Feature Extraction Methods

Hyperspectral imaging combined with feature extraction methods were applied to determine soluble sugar content (SSC) in mature and scatheless strawberry. Hyperspectral images of 154 strawberries covering the spectral range of 874～1 734 nm were captured and the spectral data were extracted from the hyperspectral images, and the spectra of 941～1 612 nm were preprocessed by moving average (MA). Nineteen samples were defined as outliers by the residual method, and the remaining 135 samples were divided into the calibration set (n=90) and the prediction set (n=45). Successive projections algorithm (SPA), genetic algorithm partial least squares (GAPLS) combined with SPA, weighted regression coefficient (Bw) and competitive adaptive reweighted sampling (CARS) were applied to select 14, 17, 24 and 25 effective wavelengths, respectively. Principal component analysis (PCA) and wavelet transform (WT) were applied to extract feature information with 20 and 58 features, respectively. PLS models were built based on the full spectra, the effective wavelengths and the features, respectively. All PLS models obtained good results. PLS models using full spectra and features extracted by WT obtained the best results with correlation coefficient of calibration (r_c) and correlation coefficient of prediction (r_p) over 0.9. The overall results indicated that hyperspectral imaging combined with feature extraction methods could be used for detection of SSC in strawberry.